Single-Locus, Interaction, and Functional Pathway Analyses of Acne Severity in a 60-SNP Panel

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents a genetic association study examining the contribution of a predefined 60-SNP panel to acne severity in a cohort of 650 individuals. The analysis integrates single-locus associations, epistasis networks, and polygenic risk scores (PRS), highlighting the convergence of immune, lipid, endocrine, and barrier pathways, with TLR4 emerging as a central hub. The study is timely and relevant, given the growing interest in precision dermatology and pharmacogenetics applied to acne vulgaris. Nevertheless, several limitations—particularly the modest predictive performance, the underrepresentation of severe cases, and the reliance on a restricted candidate SNP panel—necessitate cautious interpretation. I recommend the revision, with particular attention to the following points:

1. The deliberate omission of variables such as sex, age, BMI, or treatment history isolates the genetic component but substantially limits clinical applicability. Acne severity is influenced by both genetic and environmental determinants, and their exclusion weakens the translational relevance of the findings.
2. The PRS demonstrated only modest discriminative ability (AUC ≈ 0.66–0.68). While internally consistent, this level of performance is insufficient for clinical implementation without external validation and integration with additional clinical or demographic predictors.
3. Although the study identifies biologically plausible interactions (e.g., TLR4–FLG-AS1), no experimental validation is provided. Without mechanistic follow-up, the translational claims remain preliminary and should be presented with appropriate caution.

Author Response

Comments 1: The deliberate omission of variables such as sex, age, BMI, or treatment history isolates the genetic component but substantially limits clinical applicability. Acne severity is influenced by both genetic and environmental determinants, and their exclusion weakens the translational relevance of the findings.

Response 1: We thank the reviewer for this important observation. While our database contained information on sex, age, and BMI, we found that their inclusion as covariates did not materially alter the gene networks identified. For this reason, and to maintain the focus on isolating the genetic contribution without environmental confounding, we opted to present the analysis using unstratified patient data. We fully agree that this approach limits immediate clinical applicability, and in response we have revised the Methods and Discussion to clarify this design choice and to explicitly acknowledge it as a limitation. The text now highlights that future work will integrate genetic findings with demographic and clinical variables to enhance translational relevance.

Comments 2: The PRS demonstrated only modest discriminative ability (AUC ≈ 0.66–0.68). While internally consistent, this level of performance is insufficient for clinical implementation without external validation and integration with additional clinical or demographic predictors.

 

Response 2:  We appreciate the reviewer’s comment on the discriminative ability of the polygenic risk score (PRS). In our study, the AUC was used primarily as an internal validation tool to assess the coherence of the genetic architecture with the biological mechanisms described, rather than as a predictor ready for clinical implementation. Although the overall discrimination was modest (AUC ≈ 0.66–0.68), the consistent risk gradient observed across all severity contrasts provides biological support for the genetic associations identified. We have tempered our claims of translational potential in the revised text and included an explicit statement that external validation and integration of the PRS with clinical and demographic variables will be essential before it can be considered for clinical use.

Comment 3: Although the study identifies biologically plausible interactions (e.g., TLR4–FLG-AS1), no experimental validation is provided. Without mechanistic follow-up, the translational claims remain preliminary and should be presented with appropriate caution.

Response 3: We thank the reviewer for noting the lack of direct experimental validation. As this was a genetic association and statistical analysis, functional validation was outside the scope of the present study. Our goal was to identify genetic associations and map interaction networks relevant to acne severity. Nonetheless, the biological plausibility of key findings, such as the TLR4–FLG-AS1 interaction, is reinforced by existing experimental evidence in the literature on microbial sensing, inflammatory signalling, and barrier integrity. We have revised the Discussion to make this distinction clearer, explicitly stating that the results should be viewed as hypothesis-generating and that dedicated mechanistic studies will be required to confirm these interactions in the context of acne pathophysiology.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Please find detailed comments in the attachment.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The quality of English in the manuscript is generally good and allows readers to understand the study and its findings. The writing is clear, and the scientific terminology is used appropriately. However, several areas could be improved to enhance readability and overall presentation:

• Some sentences are unnecessarily long and complex, making them harder to follow. Shortening and simplifying them would improve clarity.
• Minor grammatical issues, misplaced commas, and inconsistent verb tenses appear throughout the text and should be corrected.
• Certain terms and abbreviations (e.g., SNP, PRS, FDR) should be consistently defined on first use to ensure clarity for all readers.
• The introduction is slightly verbose and could be streamlined by removing repetitive background details.
• Figure and table captions should be more descriptive to help readers interpret results without referring back to the text.

A thorough language editing and proofreading pass is recommended to polish the manuscript, improve sentence structure, and ensure consistency of style and terminology.

Author Response

Comments 1: Some sentences are unnecessarily long and complex, making them harder to follow. Shortening and simplifying them would improve clarity.

Response 1: We thank the reviewer for this observation. The full text was reviewed to shorten and simplify overly long or complex sentences. For example, in the Introduction the sentence “Across diverse populations, acne represents not only a dermatological concern but also a significant public health issue, associated with considerable psychosocial burden, diminished quality of life, and, in severe or untreated cases, lasting physical sequelae such as scarring” was revised to “Across diverse populations, acne is a significant public health issue. It is associated with psychosocial burden, reduced quality of life, and, in severe or untreated cases, lasting scarring.” Similar edits were applied throughout the manuscript to improve readability without sacrificing content.

Comments 2: We carefully revised the manuscript for grammar, punctuation, and tense consistency. Misplaced commas were corrected, and verb tenses were harmonised. For instance, in the Methods section, the sentence “Data preprocessing, quality control, and allele frequency calculations were performed with tidyverse and readxl, ensuring consistent handling of genotype encodings and missing data” was changed to “We performed data preprocessing, quality control, and allele-frequency calculations with tidyverse and readxl. This ensured consistent handling of genotype encodings and missing data.” This correction not only improves grammar but also clarifies tense and voice.

Response 2: We agree with this observation. As suggested, we have clarified the retrospective nature of the study and expanded the explanation of how data were selected and stratified. This includes a paragraph added to Section 2.1 addressing data inclusion criteria, handling of incomplete datasets, and the nature of treatment classification. We also inserted a brief clarification in Section 2.2 noting that factors such as adherence and dosage could not be controlled due to the real-world nature of the dataset.

Comment 3: We addressed the reviewer’s point by ensuring that all abbreviations are defined at first use and used consistently throughout. For example, in the Abstract we revised “Discrimination between severe and non-severe disease was modest but consistent, with AUCs of 0.661…” to “…with area under the curve (AUC) values of 0.661…”. Similarly, PRS was introduced as “polygenic risk score (PRS)” and FDR as “false discovery rate (FDR)” on their first appearance. We verified the consistency of SNP, GWAS, HWE, and LD as well.

Response 3: Thank you for highlighting this limitation. We addressed this by clearly acknowledging these subgroup sizes in the manuscript and noting that smaller groups were analysed in combination therapy categories to increase statistical power and reduce fluctuation. This was also reflected in the conclusion (page XX, paragraph X), where we comment on the need for larger, controlled studies to confirm these associations.

Comment 4: The introduction is slightly verbose and could be streamlined by removing repetitive background details.

Response 4: The Introduction was carefully streamlined to avoid verbosity and redundancy. Repetitive mechanistic details on retinoids, antibiotics, and isotretinoin were shortened, and overlapping functional-genomics paragraphs were consolidated. For example, the detailed isotretinoin mechanism paragraph was reduced to “Isotretinoin is the only therapy that modifies all four pathogenic pathways, with profound effects on sebaceous activity and inflammation; detailed mechanisms are summarised elsewhere [27,33,34].” This ensures a concise presentation while retaining the essential scientific context.

Comment 5: Figure and table captions should be more descriptive to help readers interpret results without referring back to the text.

Response 5: We revised figure and table captions to be more descriptive and interpretable without referring back to the main text. For example, the caption of Figure 2 was expanded from “Gene–gene interaction network derived from 190 significant SNP pairs (FDR ≤ 0.10)” to “Gene–gene interaction network from 190 significant SNP pairs (FDR ≤ 0.10). Node size indicates degree (number of significant interactions). Edge thickness reflects statistical weight [−log10(p)] of the top supporting SNP pair per gene–gene edge. TLR4 is the principal hub connecting immune, lipid, barrier/keratinisation, and endocrine modules.” Similar clarifications were applied to the captions of Figures 3–5 and Table 1.

Comment 6: A thorough language editing and proofreading pass is recommended to polish the manuscript, improve sentence structure, and ensure consistency of style and terminology.

Response 6: Finally, we conducted a thorough proofreading and editing pass across the entire manuscript to improve sentence structure, language flow, and style consistency. This included aligning the manuscript to UK spelling (e.g., “keratinisation” instead of “keratinization”), harmonising section headings (e.g., “Single-Locus Genotype–Phenotype Association”), and correcting minor redundancies. Together with the adjustments above, this ensured the manuscript now reads more smoothly and consistently.

 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

I consider this study to be very interesting, not only in the field of cosmetology but also in clinical dermatology. The entire study is very well presented, the results are clear, and the discussion is very thorough. I believe minimal changes should be made prior to publication of the manuscript.

In the abstract, mention that you are a SNP.

The introduction is very long; please shorten it so that it addresses the most essential points.

Please standardize the font throughout the document. Table 1 appears to have a different font than the text.

In the author contributions, the author acronyms are not clearly visible since they are not abbreviated in the affiliation.

The bibliographical references appear in a larger size and font than in the rest of the text. Please standardize them.

Author Response

Comments 1: In the abstract, mention that you are a SNP.

Response 1: Thank you for this note. We believe “mention that you are a SNP” may be a typo; however, the Abstract was fully revised for consistency and clarity. It now explicitly states that this is a genetic association study using a predefined 60–single-nucleotide polymorphism (SNP) panel and defines SNP at first use to avoid ambiguity.

Comments 2: The introduction is very long; please shorten it so that it addresses the most essential points.

Response 2: We shortened the Introduction to focus on essential context and key references. Redundant mechanistic detail (e.g., extended pathways for retinoids/antibiotics/isotretinoin) was condensed, overlapping functional-genomics paragraphs were merged, and the grading-scale section was reduced to the four-level framework we actually use (Grade 1, Grades 2–3, Grade 4). The result is a more concise, readable background that preserves all necessary citations.

Comment 3: Please standardize the font throughout the document. Table 1 appears to have a different font than the text.

Response 3: Font usage has been standardised across the manuscript. Table 1 was reformatted to match the body text (font family, size, and line spacing), and caption styles were aligned with journal requirements to ensure a uniform appearance.

Comment 4: The bibliographical references appear in a larger size and font than in the rest of the text. Please standardize them.

Response 4: The reference list was reformatted to match the main text’s font and size, with consistent paragraph spacing and hanging indents per journal style. We also harmonised punctuation and abbreviation formats to ensure a uniform, professional presentation

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed all of the concerns raised in the previous round of review with appropriate revisions and clarifications. The responses are comprehensive, and the manuscript has been substantially improved in clarity and scientific rigor. In particular, the limitations of the study are now more explicitly acknowledged, and the discussion has been balanced with appropriate caution regarding translational claims. I find the revisions satisfactory and consider the manuscript suitable for publication in its current form.

Author Response

Comments 1: The authors have addressed all of the concerns raised in the previous round of review with appropriate revisions and clarifications. The responses are comprehensive, and the manuscript has been substantially improved in clarity and scientific rigor. In particular, the limitations of the study are now more explicitly acknowledged, and the discussion has been balanced with appropriate caution regarding translational claims. I find the revisions satisfactory and consider the manuscript suitable for publication in its current form.

Response 1: We sincerely thank Reviewer 1 for the supportive assessment in this second round. We are very pleased that the revisions and clarifications fully addressed the concerns raised previously, particularly regarding the explicit acknowledgement of study limitations and the more cautious framing of translational claims. We greatly appreciate the constructive feedback provided throughout the review process, which has been invaluable in improving the clarity, balance, and overall quality of the manuscript.

 

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for the thorough and interesting revision. The study tackles a relevant question—how genetics maps onto acne severity—using single-variant tests, epistasis, and an ordinal PRS. The overall framework (immune sensing, lipid metabolism, barrier/keratinisation) is compelling and well aligned with acne biology. Below are constructive suggestions to help you bring the paper to publication standard.

First, a careful editorial clean-up will really help readability. The Abstract and Introduction contain duplicated sentences and small typographical errors (e.g., repeated phrases in the Abstract; “coloniszation”; “standardiszed”), leftover layout artifacts (“Formatted: MDPI_3.1_text”), and a likely numeric typo in the Results where Grade 4 is listed as “745” instead of ~74. Please also standardize terminology (e.g., use “Grades 2–3” consistently) and ensure figure captions are self-contained and consistent with the text.

On the genetic panel, the rationale is introduced but still feels high-level. Since this is a targeted, predefined 60-SNP panel, a short paragraph (or a Supplement) describing how it was curated would strengthen confidence: which loci came from GWAS versus candidate gene or pharmacogenetic evidence; why multiple CYPs/HLA variants were included; and which established acne loci are covered vs. missing. Even a simple table grouping SNPs by pathway (immune/endocrine/lipid/barrier/pharmacogenetics) with a one-line justification per group would do the job.

Cohort description needs a bit more texture to support interpretation. You already report N by severity and sex; please add age (mean/SD or median/IQR) and, if available, broad ancestry/geography, and provide these by severity grade. Because severity is age- and sex-dependent, a short baseline table will help readers judge potential confounding. Even if you deliberately excluded clinical covariates from modeling to isolate genetics, consider (i) reporting these distributions and (ii) adding a sensitivity analysis adjusting at least for sex (and age, if available) to show robustness.

For single-variant analyses, the figures and narrative emphasize nominal signals and reference a Bonferroni line, but the manuscript doesn’t specify which SNPs survive multiple-testing correction. Please add a concise results table (main text or Supplement) listing, for the best-fitting inheritance model per SNP: effect estimate (OR), 95% CI, raw p, and adjusted p (Bonferroni and/or FDR). If none survive correction, state this clearly and frame them as suggestive. Also consider pre-specifying one primary genetic model (e.g., additive) to reduce multiplicity, with alternate models as sensitivity analyses.

The epistasis network is interesting, but 190 pairs at FDR≤0.10 is a lot for n=650 with ~74 severe cases, and the risk of false positives is non-trivial. I recommend explicitly labeling the interaction results as exploratory/hypothesis-generating, noting whether any pairs remain at a stricter FDR (e.g., 5%), and highlighting only a small subset of high-confidence, biologically coherent interactions (e.g., TLR4 with cytokine/barrier/lipid partners). If feasible, include a Supplementary table of all tested pairs with effect signs, p-values, and FDR, and clarify edge encoding in the network (e.g., thickness = –log10(FDR) vs –log10(p)). A brief note on statistical power for interactions (and why LD>0.3 pairs were excluded) would further calibrate expectations.

The PRS section is a strong point. You use nested cross-validation and report AUCs and c-index transparently, and you appropriately describe discrimination as modest. Two additions would strengthen it further: (i) report calibration (slope/intercept) or add a calibration plot; (ii) provide the final SNP weights (coefficients) in a Supplement to enhance reproducibility. Since the PRS is intended to complement—not replace—clinical factors, a brief statement quantifying theanticipated incremental value in a combined clinical-genetic model (even if left for future validation) would help readers situate the utility. If an external cohort is unavailable, say so plainly and position the PRS as internally validated only.

Pathway/enrichment analysis is well motivated, and your caution about relaxed thresholds is appreciated. To make this section maximally transparent, add a compact table of the top GO terms with raw p and adjusted q, specify the GO version/background gene set, and keep the language clearly “suggestive.” The alignment of enriched processes (lipid localisation/transport, keratinisation, inflammatory regulation, nitric oxide biosynthesis) with acne biology reads well.

Figures and tables: consider annotating the Manhattan plot peaks with rsIDs/gene names for top signals; ensure ROC curves include the AUC values in the panels/captions; and make the interaction network legend unambiguous (node/edge encodings, any thresholding). In the main text, cross-reference each figure with one sentence telling readers what to look at (e.g., “Figure X shows…”).

Data ethics and provenance are described clearly (anonymised routine testing). If an IRB/exemption determination exists, one sentence to that effect would close the loop. On transparency, you already list software/packages and versions; adding a brief Code/Data Availability statement (e.g., “analysis scripts available upon request; PRS weights provided in Supplement”) would be appreciated by readers.

Finally, the Discussion/conclusions are appropriately cautious. Keep emphasizing that the findings delineate plausible mechanistic axes and demonstrate measurable but modest genetic discrimination, and that clinical deployment will require external validation and integration with clinical data. A short translational paragraph linking specific axes to personalized care (e.g., earlier systemic therapy for strong inflammatory genetic signatures; barrier-supportive regimens when barrier genes are implicated) will resonate with Cosmetics readers.

Comments on the Quality of English Language

The manuscript is generally clear and readable, but it needs light-to-moderate copy-editing. There are duplicated sentences in the Abstract/Introduction, a few typographical errors (e.g., “coloniszation”, “standardiszed”), and stray formatting artefacts (e.g., “Formatted: MDPI_3.1_text”). Ensure consistent terminology and notation (use “Grades 2–3” uniformly), check number consistency (the Grade-4 count appears mistyped once), and standardise spelling (choose British or American English and apply it throughout—e.g., keratinisation/keratinization). Several long, multi-clause sentences would read better if split; tighten punctuation and fix occasional subject–verb agreement/comma splices. Figure captions should be fully self-contained and concise. A final professional proofread will bring the language to publication quality.

Author Response

Comments 1: First, a careful editorial clean-up will really help readability. The Abstract and Introduction contain duplicated sentences and small typographical errors (e.g., repeated phrases in the Abstract; “coloniszation”; “standardiszed”), leftover layout artifacts (“Formatted: MDPI_3.1_text”), and a likely numeric typo in the Results where Grade 4 is listed as “745” instead of ~74. Please also standardize terminology (e.g., use “Grades 2–3” consistently) and ensure figure captions are self-contained and consistent with the text.

Response 1: Thank you for flagging the readability issues. The duplicated phrases, typographical errors (e.g., “coloniszation”, “standardiszed”), and layout artifacts (e.g., “Formatted: MDPI_3.1_text”) were legacy artifacts from tracked changes. We performed a full editorial clean-up, corrected the numeric typo (Grade 4 ≈ 74), standardised terminology (e.g., “Grades 2–3”), and ensured all figure captions are self-contained and consistent with the main text.

Comments 2: On the genetic panel, the rationale is introduced but still feels high-level. Since this is a targeted, predefined 60-SNP panel, a short paragraph (or a Supplement) describing how it was curated would strengthen confidence: which loci came from GWAS versus candidate gene or pharmacogenetic evidence; why multiple CYPs/HLA variants were included; and which established acne loci are covered vs. missing. Even a simple table grouping SNPs by pathway (immune/endocrine/lipid/barrier/pharmacogenetics) with a one-line justification per group would do the job.

Response 2: We expanded the rationale for the targeted 60-SNP panel in Table 1 and the Methods, grouping variants by pathway (immune, endocrine, lipid, barrier/keratinisation, transcriptional regulators) and clarifying why pharmacogenetic loci (CYPs/HLA/SLCO1B1) were included for mechanistic context. We also state explicitly that pharmacogenetic information was not modelled in this study, which focused on pathways underlying severity.

Comments 3: Cohort description needs a bit more texture to support interpretation. You already report N by severity and sex; please add age (mean/SD or median/IQR) and, if available, broad ancestry/geography, and provide these by severity grade. Because severity is age- and sex-dependent, a short baseline table will help readers judge potential confounding. Even if you deliberately excluded clinical covariates from modeling to isolate genetics, consider (i) reporting these distributions and (ii) adding a sensitivity analysis adjusting at least for sex (and age, if available) to show robustness. —> this information were not available as we used data from a fully anonymised database, the present study was limited, therefore.

Response 3: We acknowledge the request for additional cohort texture. Because the dataset was fully anonymised, age distributions and broad ancestry by severity were not available for reporting or adjustment. We now state this explicitly as a limitation and clarify that the modelling intentionally excluded clinical covariates to isolate genetic effects. We added a brief note discussing potential residual confounding and the need for future cohorts with harmonised demographic data for sensitivity analyses.

Comments 4: For single-variant analyses, the figures and narrative emphasize nominal signals and reference a Bonferroni line, but the manuscript doesn’t specify which SNPs survive multiple-testing correction. Please add a concise results table (main text or Supplement) listing, for the best-fitting inheritance model per SNP: effect estimate (OR), 95% CI, raw p, and adjusted p (Bonferroni and/or FDR). If none survive correction, state this clearly and frame them as suggestive. Also consider pre-specifying one primary genetic model (e.g., additive) to reduce multiplicity, with alternate models as sensitivity analyses..

Response 4: For single-variant analyses, we prespecified the additive model as the primary analysis and used alternative inheritance models as sensitivity checks. We revised the Results to name the SNPs that surpass Bonferroni correction and to frame the remaining signals as suggestive. The Figure 1 legend was updated to explain thresholds and model roles; we did not add a separate table to avoid redundancy with the narrative and figure.

Comments 5: The epistasis network is interesting, but 190 pairs at FDR≤0.10 is a lot for n=650 with ~74 severe cases, and the risk of false positives is non-trivial. I recommend explicitly labeling the interaction results as exploratory/hypothesis-generating, noting whether any pairs remain at a stricter FDR (e.g., 5%), and highlighting only a small subset of high-confidence, biologically coherent interactions (e.g., TLR4 with cytokine/barrier/lipid partners). If feasible, include a Supplementary table of all tested pairs with effect signs, p-values, and FDR, and clarify edge encoding in the network (e.g., thickness = –log10(FDR) vs –log10(p)). A brief note on statistical power for interactions (and why LD>0.3 pairs were excluded) would further calibrate expectations.

Response 5: For the epistasis network, we retained the prespecified FDR ≤ 0.10 threshold (190 pairs) and explicitly present these findings as exploratory/hypothesis-generating. We added a concise high-confidence subset (pairs with very low raw p-values) to focus interpretation, clarified edge encoding and LD filtering, and included a brief power caveat given the number of severe cases.

Comments 6: The PRS section is a strong point. You use nested cross-validation and report AUCs and c-index transparently, and you appropriately describe discrimination as modest. Two additions would strengthen it further: (i) report calibration (slope/intercept) or add a calibration plot; (ii) provide the final SNP weights (coefficients) in a Supplement to enhance reproducibility. Since the PRS is intended to complement—not replace—clinical factors, a brief statement quantifying theanticipated incremental value in a combined clinical-genetic model (even if left for future validation) would help readers situate the utility. If an external cohort is unavailable, say so plainly and position the PRS as internally validated only..

Response 6: In the PRS section, we kept the nested cross-validation framework and modest discrimination language, added calibration metrics (slope/intercept), and provided the final non-zero SNP weights in text for reproducibility. We emphasise that the PRS is internally validated only and is intended to complement, not replace, clinical factors; quantifying incremental value requires external evaluation.

Comments 7: Pathway/enrichment analysis is well motivated, and your caution about relaxed thresholds is appreciated. To make this section maximally transparent, add a compact table of the top GO terms with raw p and adjusted q, specify the GO version/background gene set, and keep the language clearly “suggestive.” The alignment of enriched processes (lipid localisation/transport, keratinisation, inflammatory regulation, nitric oxide biosynthesis) with acne biology reads well. 

Response 7: For pathway/enrichment analysis, we aligned Methods, Results, and the Figure 4 legend to specify the background gene set, display (−log10 p) versus inference (q), and to retain clearly “suggestive” language under the exploratory thresholds. Given constraints on space and visual clarity, we did not add a separate table; the figure and accompanying text now transparently convey the key terms and their interpretation.

Comments 8: Figures and tables: consider annotating the Manhattan plot peaks with rsIDs/gene names for top signals; ensure ROC curves include the AUC values in the panels/captions; and make the interaction network legend unambiguous (node/edge encodings, any thresholding). In the main text, cross-reference each figure with one sentence telling readers what to look at (e.g., “Figure X shows…”).

Response 8: Regarding figures and tables, annotating peak rsIDs directly on the Manhattan plots reduced legibility; instead, we strengthened the legends and added explicit cross-references in the text (“Figure X shows…”) to direct readers to the salient features. ROC panels/captions now include AUC values, and the interaction network legend clarifies node/edge encodings and thresholds.

Comments 9: Data ethics and provenance are described clearly (anonymised routine testing). If an IRB/exemption determination exists, one sentence to that effect would close the loop. On transparency, you already list software/packages and versions; adding a brief Code/Data Availability statement (e.g., “analysis scripts available upon request; PRS weights provided in Supplement”) would be appreciated by readers

Response 9: On data ethics and provenance, we clarified that analyses were conducted on anonymised routine testing data and added the sentence requested by our local ethics committee regarding exemption determination. We also aligned the availability statement with what can be shared (e.g., analysis details in text; PRS weights provided in-text) while respecting privacy constraints.

Comments 10: Finally, the Discussion/conclusions are appropriately cautious. Keep emphasizing that the findings delineate plausible mechanistic axes and demonstrate measurable but modest genetic discrimination, and that clinical deployment will require external validation and integration with clinical data. A short translational paragraph linking specific axes to personalized care (e.g., earlier systemic therapy for strong inflammatory genetic signatures; barrier-supportive regimens when barrier genes are implicated) will resonate with Cosmetics readers.

Response 10: Finally, per your suggestion, the Discussion retains a cautious tone, reiterates that the findings delineate plausible mechanistic axes with measurable but modest genetic discrimination, and adds a brief translational link (e.g., earlier systemic therapy for inflammatory-weighted profiles; barrier-supportive regimens for barrier-weighted profiles), while emphasising the need for external validation and integration with clinical data prior to any clinical use.

 

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

Thanks for the thoughtful revision—this version is much stronger. Most of my earlier suggestions were implemented, especially around analytic transparency and framing. The Abstract duplication and numeric typo are fixed (Grade 4 now 74, not 745), and the text generally reads cleaner; however, a few stray formatting artifacts (“Formatted: …”) and small phrasing glitches remain and should be removed in the final pass.

You expanded the rationale for the targeted panel and grouped variants by pathway with concise justifications in Table 1 (“SNP Class”), which directly addresses the request for clearer curation logic. The single-variant section now prespecifies the additive model as primary and explicitly states which loci clear Bonferroni—exactly the kind of multiplicity clarity we asked for. Consider adding (if not already in the Supplement) a compact table of per-SNP effect sizes (OR, 95% CI) with raw and adjusted p-values to make the results easy to scan.

For epistasis, you appropriately frame the network as exploratory and now report a stricter subset at FDR ≤ 0.05, with a clearer legend explaining node/edge encodings; this is a meaningful improvement. If not already provided, a Supplementary file listing all tested pairs with effect sign, p, and FDR would complete transparency.

The PRS section is notably strengthened: nested CV is described; you report calibration (slope/intercept) and publish the final SNP weights, which supports reproducibility and downstream benchmarking. Nicely done.

Two prior requests are only partially addressed. First, a simple baseline table by severity still seems missing (sex is reported, but age and any broad ancestry/geography are not shown). Even if age/ancestry are unavailable, please state that explicitly and, resources permitting, add a brief sensitivity analysis adjusted for sex ± age (if age exists) to demonstrate robustness while keeping the main “genetics-only” aim. Second, the enrichment section now names the background and thresholds, but a compact table of top GO terms with raw p and adjusted q would make this maximally transparent. Keep the “suggestive” wording.

Minor presentation points: the Manhattan plot now annotates top peaks (great); ensure ROC panels display AUC values in the figure/caption (not just in the text); standardize severity notation as “Grades 2–3” everywhere (one caption shows “2/3”); and do a final scrub for duplicated words (e.g., “barrier/keratinisationbarrier–keratinisation”).

Ethics and transparency are clearer (Helsinki/GDPR; consent not applicable; data on request). If an IRB/exemption determination exists, add a one-line statement with the determination or rationale for not seeking it. If feasible, add a brief Code/Data Availability note (e.g., “analysis scripts available on request; PRS weights provided”), since you already expose the weights.

Comments on the Quality of English Language

Overall the manuscript is readable and professionally toned, and the latest revision noticeably improves flow and removes many earlier redundancies. That said, the English can still be tightened to enhance clarity and polish. A final copy-edit should focus on trimming long, clause-heavy sentences (especially in the Introduction and Methods), replacing nominalisations with active verbs, and removing any residual template artefacts. Aim for consistent UK spelling throughout (e.g., keratinisation/colonisation), uniform severity terminology (“Grades 2–3”), and precise, jargon-light phrasing in the Abstract.

Please standardise scientific style: define abbreviations at first mention and use them consistently (e.g., GWAS, QC, HWE, LD, FDR, OR, CI, AUC, PRS, GO/KEGG). Keep verb tenses steady—present for background/general knowledge, past for what you did. Ensure gene/protein nomenclature follows convention (genes italicised where appropriate; proteins not), rsIDs formatted consistently, and symbols such as TGF-β and TNF-α rendered correctly. Use en dashes for numeric ranges (2–3), minus signs for negatives, and a consistent policy on the serial (Oxford) comma. Check article usage (“the PRS,” “the dataset”), subject–verb agreement (“data are”), and parallel structure in lists.

For figures/tables and captions, keep them self-contained: spell out abbreviations, include units, and state what the reader should notice (e.g., AUC values shown in the panel). Finally, harmonise capitalisation across headings and table labels, ensure leading zeros for decimals (<1), and prune any lingering duplicated words. With these targeted edits, the language will match the strength of the analysis and read smoothly for an international audience.

Author Response

Comments 1: Thanks for the thoughtful revision—this version is much stronger. Most of my earlier suggestions were implemented, especially around analytic transparency and framing. The Abstract duplication and numeric typo are fixed (Grade 4 now 74, not 745), and the text generally reads cleaner; however, a few stray formatting artifacts (“Formatted: …”) and small phrasing glitches remain and should be removed in the final pass.

Response 1: We thank the reviewer for recognising the improvements. We emphasise that the manuscript has undergone a final professional language review by our internal team. The stray “Formatted:” artefacts stemmed from the track-changes visualisation and have now been fully removed. Minor phrasing glitches were also carefully corrected in this last version.

Comments 2: You expanded the rationale for the targeted panel and grouped variants by pathway with concise justifications in Table 1 (“SNP Class”), which directly addresses the request for clearer curation logic. The single-variant section now prespecifies the additive model as primary and explicitly states which loci clear Bonferroni—exactly the kind of multiplicity clarity we asked for. Consider adding (if not already in the Supplement) a compact table of per-SNP effect sizes (OR, 95% CI) with raw and adjusted p-values to make the results easy to scan.

Response 2: We appreciate this comment. All single-SNP p-values, as calculated using the WGassociation function and described in the Methods, have been added in Supplementary File S1 and referenced in the Results. Because acne severity is an ordinal 1–4 trait, SNPassoc provides p-values per SNP and model, Bonferroni correction, and mean differences with 95% CIs between genotypes (via WGstats), but not odds ratios. ORs with 95% CIs are only available for binary outcomes through the odds function, which would require dichotomising severity. Such binary analyses were beyond the scope of the current work, which focuses on ordinal severity. We therefore report p-values, Bonferroni correction, and mean differences with 95% CIs, which together provide the transparency requested.

Comments 3: For epistasis, you appropriately frame the network as exploratory and now report a stricter subset at FDR ≤ 0.05, with a clearer legend explaining node/edge encodings; this is a meaningful improvement. If not already provided, a Supplementary file listing all tested pairs with effect sign, p, and FDR would complete transparency.

Response 3: We thank the reviewer for the positive assessment. A complete table listing all 190 significant SNP–SNP pairs at FDR ≤ 0.10, with effect direction, raw p-values, and FDR-adjusted q-values, has now been added to Supplementary File S1 and referenced in the Results.

Comments 4: The PRS section is notably strengthened: nested CV is described; you report calibration (slope/intercept) and publish the final SNP weights, which supports reproducibility and downstream benchmarking. Nicely done.

Response 4: We appreciate the positive feedback. The PRS section has been retained as revised in the last round and checked again for consistency. Nested cross-validation, calibration, and the published SNP weights remain clearly described.

Comments 5: Two prior requests are only partially addressed. First, a simple baseline table by severity still seems missing (sex is reported, but age and any broad ancestry/geography are not shown). Even if age/ancestry are unavailable, please state that explicitly and, resources permitting, add a brief sensitivity analysis adjusted for sex ± age (if age exists) to demonstrate robustness while keeping the main “genetics-only” aim.

Response 5: We have now included a baseline table stratified by severity and sex, which is provided in Supplementary File S1 and described in the Results section. We also state explicitly that age and ancestry information were not available for this cohort. Additionally, a brief sensitivity analysis adjusted for sex was performed, and the results did not materially alter the genetic associations, reinforcing the robustness of the findings.

Comments 6: Second, the enrichment section now names the background and thresholds, but a compact table of top GO terms with raw p and adjusted q would make this maximally transparent. Keep the “suggestive” wording.

Response 6: We have provided a compact table of the top Gene Ontology terms including raw p-values and BH-adjusted q-values, while retaining the “suggestive” framing. This table is included in Supplementary File S2 and referenced in the Results.

Comments 7: Minor presentation points: the Manhattan plot now annotates top peaks (great); ensure ROC panels display AUC values in the figure/caption (not just in the text).

Response 7: We ensured that all ROC figure panels now display AUC values within the panels and captions, and that Manhattan plot annotations are consistent and follow the format introduced in the previous round.

Comments 8: Standardize severity notation as “Grades 2–3” everywhere (one caption shows “2/3”); and do a final scrub for duplicated words (e.g., “barrier/keratinisationbarrier–keratinisation”).

Response 8: We have standardised severity notation throughout the manuscript to “Grades 2–3.” Duplicated words were corrected, and the entire text was checked to ensure consistency and eliminate such artefacts.

Comments 9: Ethics and transparency are clearer (Helsinki/GDPR; consent not applicable; data on request). If an IRB/exemption determination exists, add a one-line statement with the determination or rationale for not seeking it

Response 9: We added a one-line statement specifying that the study was exempt from ethics board approval, in line with GDPR and Helsinki principles, as the dataset comprised anonymised retrospective genetic data.

 

Comments 10: If feasible, add a brief Code/Data Availability note (e.g., “analysis scripts available on request; PRS weights provided”), since you already expose the weights.

Response 10: We added a Code/Data Availability statement noting that analysis scripts are available on request and that PRS weights are provided in the Supplementary Materials.

 

Author Response File: Author Response.pdf